Blinded independent radiologic review (BIRR) versus investigator radiologic evaluation (IRE) for response assessment in metastatic colorectal cancer (mCRC): Systematic review and meta-analysis.

Abstract

2559 Background: Randomized controlled trials (RCT) provide the most reliable evidence about therapeutic interventions. However, RCTs are still in risk of some bias, like assessment bias. In order to reduce assessment bias, BIRR has been frequently used in clinical research. Despite the widespread use of BIRR, its real impact on bias reduction is fairly unknown. We performed a systematic review and meta-analysis to evaluate the impact of BIRR on response evaluation in mCRC. Methods: RCT testing any drug therapy in mCRC was eligible if overall response rate (ORR) data from BIRR and IRE were presented. Medline, Embase, CENTRAL, Clinicaltrials.gov, ASCO, and ESMO proceedings were searched. Random-effects model were used in calculations. Heterogeneity was expressed as I². Risk ratio (RR) and 95% confidence intervals (95%CI) were computed for BIRR and IRE data. Interaction tests were used to compare BIRR and IRE results. Ratios between response rates in experimental arms in BIRR and IRE (experRR) and control arms in BIRR and IRE (contRR) were also calculated. Results: 12 RCT (5189 patients) provided ORR data. Overall response rate RR in BIRR was 1.17 (95% CI 1.10, 1.25) while overall response rate RR in IRE was 1.19 (95% CI 1.11, 1.27). Interaction test showed no difference in ORR assessments between BIRR and IRE (P=0.17). The experRR (RR 0.88; 95%CI 0.79, 0.99) and contRR (RR 0.86; 95%CI 0.76, 0.97) showed that BIRR decreased the frequency of responses in experimental and control arms in a similar fashion, differently from previously hypothesized (Dodd et al., Biometrics 2011). Conclusions: In this large sample of trials, BIRR seemed to reduce the response rate in a similar amount in both control and experimental arms. These results suggest that BIRR may not improve reliability or assessment bias of randomized controlled trials. However, BIRR may be important in studies whose endpoints are response rates, such as phase II trials. Any impact of BIRR in local radiologist’s behavior, like “audit bias”, could not be evaluated and a prospective study is mandatory.